基于带噪声数据集的强鲁棒性隐含三元组质检算法

Strong Robustness Implied Triplet Quality Inspection Algorithm Based on Noisy Dataset

知识图谱三元组质检的难点是区分真实三元组和噪声三元组,常用开源知识图谱不包含噪声三元组,目前已有三元组质检算法极少考虑到知识图谱中由于关系传递存在的大量隐含三元组对质检效果的影响,且没有有效利用实体之间的空间语义关联导致对实体特征提取不充分。针对以上问题,提出带噪声数据集的强鲁棒性隐含三元组质检算法(Implied triplet quality inspection,ITQI),首先基于开源数据集制作Neo4J知识图谱;然后基于有向图最长路径搜索算法搜索所有可能的搜索路径,根据知识图谱的关系传递性来构建具有隐含关系的三元组,对源三元组进行扩充能够极大增加有效三元组的个数;最后通过随机采样构建三种类型的噪声三元组。采用TransR预训练得到扩充后的真实三元组的初始特征,然后使用残差网络提取三元组的静态特征、并使用多层BiLSTM提取三元组的内部关联特征,将以上三种特征聚合,得到三元组的融合特征对三元组进行二分类达到三元组质检的目的。论文算法在FB15K数据集上进行实验,实验结果表明论文算法质检效果优于对比算法且鲁棒性最强。

The difficulty of knowledge graph triplet quality inspection is to distinguish between real triples and noise triples,commonly used open source knowledge graphs do not contain noise triples,at present,there are triples quality inspection algorithms that rarely consider the impact of a large number of implicit triples on the quality inspection effect due to relationship transmission in the knowledge graph,and do not effectively use the spatial semantic association between entities to lead to insufficient extraction of entity features.To solve the above problems,a robust triplet quality inspection(ITQI)algorithm with noisy dataset is proposed,and Neo4J knowledge graph is first made based on open source dataset.Then,based on the longest path search algorithm of the directed graph,all possible search paths are searched,and triples with implicit relationships are constructed according to the relational transitivity of the knowledge graph,and the expansion of the source triples can greatly increase the number of effective triples.Finally,three types of noise triples are constructed by random sampling.TransR pre-training is used to extract the static features of the triplet,and then the residual network is used to extract the static features of the triplet,and the internal association features of the triplet are extracted by using multi-layer BiLSTM,and the above three features are aggregated to obtain the fusion features of the triplet and classify the triplet to achieve the purpose of triplet quality inspection.The experimental results show that the quality inspection effect of the proposed algorithm is better than that of the comparison algorithm and has the strongest robustness.